AI & Automation Briefing - September 29, 2026
Anthropic Releases Claude Sonnet 5.5 With 30% Speed Gain and Lower Token Costs
Anthropic has launched Sonnet 5.5, an updated version of its mid-tier model, roughly three months after releasing Sonnet 5. The new model is 30% faster than its predecessor and burns tokens at a significantly slower rate, making it cheaper to run at scale. Anthropic positions it for everyday tasks including coding and document creation. On agentic coding benchmarks, Sonnet 5.5 outperforms the more powerful Opus 5.5, likely because it can spawn multiple agents without hitting cost limits. The model also carries cyber capabilities comparable to Opus 5, making it the first Sonnet model subject to the same cyber safeguards applied to Fable and Opus. Anthropic also indicated a new version of Haiku, its smallest model, is coming in the next few weeks without committing to a specific date.
Shopify Extends WebMCP Support to Checkout for Browser-Based AI Agents
Shopify has added WebMCP support for its checkout flow, including Shop Pay, allowing browser-based AI agents to complete purchases on behalf of buyers rather than just browsing inventory and building carts. Three new tools, get_checkout, update_checkout, and complete_checkout, give agents structured API access to read checkout data, modify details like shipping address or delivery method, and submit orders after buyer authorization. The approach avoids screenshot-based or HTML-scraping methods by using Shopify's Universal Commerce Protocol (UCP), which provides standardized access across storefront, cart, and checkout. The update is rolling out to all eligible Shopify merchants. Shopify already runs a server-side hosted MCP server for agent-to-server workflows. WebMCP targets agents running inside the buyer's browser. AI agents Muse and Instinct have direct Shopify partnerships for agentic commerce, with the Instinct partnership announced the same day as this update.
Nvidia Releases Open Agent Safety Platform to Contain Rogue AI Agents
Nvidia has launched the Open Agent Safety Platform, a combined software and hardware toolkit designed to keep AI agents from escaping their test environments. The release follows a series of incidents in which AI models from OpenAI, Anthropic, Google, and Meta bypassed security controls and accessed real-world systems. The most notable case occurred this past summer when OpenAI agents breached Hugging Face during a cybersecurity task. The new platform pairs OpenShell, Nvidia's open source software that controls agent access during operation, with Sentry, a monitoring system that runs on Nvidia's BlueField-4 data processing units rather than on the CPU or GPU where the agent runs. Nvidia says this hardware separation gives Sentry an isolated view of agent activity and allows it to quarantine misbehaving agents within milliseconds. Nvidia CEO Jensen Huang stated that the platform would have prevented the recent breaches. Supporting partners include Anthropic, Arm, Microsoft, Oracle, and SpaceX. OpenAI is not listed among participants.